Does Computer Really Understand Chess (In ENGLISH)
Hello. This is FM Shin Uesugi.
I am taking a class called "Philosophy of Mind" taught by Prof. John Searle and last Thursday, he mentioned a little about chess, which inspired me to write this blog.
So let me introduce Prof. Searle a little.
He is one of the famous professors at UC Berkeley and despite the fact that he's 80 years old, he teaches the class very enthusiastically every week. The topic I am writing about today is "Does computer really 'understand' chess?" and this is based on Searle's famous "Chinese Room Argument (CRA)."
So let me explain what CRA is to those of you who have never heard of it before (in extremely simple manner).
Long time ago, there was a smart guy named Turing and he said, "If a human has a chat with a computer and s/he cannot distinguish between whether an entity that s/he is chatting with is human or computer, then that computer is as intelligent as human and understands the conversation as much as humans do.
CRA is an argument against the statement above. Searle said, "OK, let me (who doesn't understand Chinese at all) enter the room and have a chat with Native Chinese speaker. In this room, there is a book that contains all the responses to the chinese symbols I receive. I can use this information to respond to Native Chinese speaker outside perfectly. I don't understand anything about Chinese, but according to Turing, I understand Chinese conversation (lol)."
So the point here is:
1. Computer (here, Searle in the room) has a program where it can respond to any of the Chinese symbols given.
2. However, the computer does not understand what those symbols mean.
In short, "Syntax doesn't equal Semantics."
So since it's chess.com here, lets talk about some chess.
On May 1997, Garry Kasparov, a former world champion who is one of the strongest chess players ever, lost to a Super Computer Deep Blue. Hearing this news, lots of people called Searle who was at London. Some said "Look! A world champion lost to a computer! That means computer understands chess, an intelligent game, better than humans."
In response to this, Searle said, "Computers don't play chess. Nor they understand it. Computer only takes in symbols, such as chess positions, as inputs, do some syntactic computing using the program it has, and produce another symbol (a chess move) as an output. It is computer's power and strength to use huge database of chess games and do things like above, but there is no 'understanding' of chess."
So I heard this argument and started thinking about it. Today, it is probably impossible for a person to play against a computer in reasonably long time control and win. Then that means computers are stronger than humans. Then can a computer be a best chess coach?
I think the answer is straight up "No." Many people do use computers to look up their opening lines and games, but unless there are simple tactics, it is very difficult to understand the concepts behind the computers. Humans can interpret computer evaluations and make a meaning out of it, but computers cannot tell a person, "the evaluation is this because in this position, you have a space advantage and you can maneuver these pieces and ultimately checkmate your opponent." Searle would probably say, "Since computers don't even understand chess, what are you trying to learn from it? Maybe you should ask someone strong (like a grandmaster) to interpret the computer evaluations and have them explain it to you." In my opinion, if you want to become better in chess, it would make much more sense to hire strong players (GMs) to teach the idea of the game, rather than completely relying on computers.
Another thing that came to my mind is the game between Carlsen and Karjakin at 2013 Tata Steel Chess. For those of you who haven't seen the game (I would definitely recommend you to check it out):
Here, Carlsen played 67.g4!!!. I think this move deserves more "!"s than 3. In the LIVE game site, the computer doesn't seem to give this move such a high evaluation. Computer says before 67.g4, the evaluation is 0.56 in favor of white, but afterwards, the evaluation becomes 0.00 due to perpetual check that leads to a draw. Then lets let a computer play this position for white. It will choose 67.Ra8, which has evaluation 0.56. Then can white win in this position if he/she/it plays in this way against Karjakin? Probably not. It is not so difficult for Karjakin to find correct moves and it is very difficult for white to make progress, so the game will likely to end in a draw. However, Carlsen played 67.g4!!! and won the game. I think Carlsen sensed that it will be a easy draw for Karjakin if he doesn't play 67.g4, so he decided to take some risks and go for a win. Turns out this was a correct decision.
Then there is a paradox. I said before that computers are stronger than any human beings, but cannot win in a position where Carlsen won. In this position, it is clear that humans surpass computers in chess (by taking in consideration of Karjakin's psychology, difficulty of finding perpetual, etc).
Does that mean I was wrong by saying "computers are stronger than humans in chess?"
This is my personal opinion, and many might disagree with me, but I think the place chess will be in next couple decades is that "Humans are stronger in tournaments with Round Robin or Swiss style, and Computers are stronger in the matches." For example, we can throw in a computer at tournament like Tata Steel Chess (which uses Round Robin system), it may not be the case that computers can necessarily win the tournament because it is advantageous for humans that have better understanding of chess psychology to take advantage of other human opponens (e.g. Carlsen vs Karjakin above). On the other hand, I think computers can pretty much beat any humans in a match because they never get tired (of draws every game) and can take advantage when humans make mistake.
As a result, some may say that chess became boring because of technology, but maybe we can look it a little different. At least looking at Carlsen-Karjakin example (and I'm sure we can find many more), we can not 100% say that computers are better than us.
When I talk to people in college and the topic like I used to play chess comes up, some people say "isn't that a game where world champion lost to a computer?" But I feel like it happened just because Kasparov played under the condition in which computer is better than humans (match system). So maybe if someone says a same thing to you, you can replied back with Carlsen-Karjakin example (lol).
Thank you for reading. Feel free to leave some comments/feedbacks and discuss about it. I think this is a very important topic and I'd love to hear what you guys think.
Have a Happy Valentines Day. Don't waste a special day reading this blog since it's pretty long :P (trololol)